fill_nan is not a valid aggregation in group_by

[paladin158]

Checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of Polars.

Reproducible example

import polars as pl

df = pl.DataFrame({"a": [1, 1, 2, 2], "b": [1, 2, 0, 0]})

df.group_by("a").agg((pl.col("b") / pl.col("b").std()).fill_nan(0).sum())

Log output

The predicate '[(col("b").cast(Float64)) / (col("b").std())].is_not_nan()' in 'when->then->otherwise' is not a valid aggregation and might product a different number of rows than the group_by operation would. This behavior is experimental and may be subject to change

Issue description

I don't quite get why I receive such a warning message. Although the results are correct, I am not sure about the implications of such a warning.

Expected behavior

No warning message.

Installed versions

0.19.3

[taozuoqiao]

I would avoid when->then->otherwise related expressions in the agg context, and prefer something like

(
    df.with_columns(
        pl.col("b").truediv(pl.col("b").std().over("a")).fill_nan(0).alias("b")
    )
    .group_by("a")
    .agg(pl.col("b").sum())
)

[stinodego]

Related to #10055

fill_nan does a when->then->otherwise behind the screens, apparently.